Native Hybrid AI
AI that fuses machine learning and operations research into one method, solving industrial problems that neither field can handle alone.
Most AI solutions that use machine learning (ML) and operations research (OR) chain them: ML predicts, then OR optimizes on those predictions. Funartech goes further. We design new algorithms in which ML, OR and, when needed, other mathematical sciences are fused at a fundamental level and work as a single method rather than as separate black boxes.
Native Hybrid AI turns predictions directly into optimal decisions. It stays robust when data is scarce or noisy, and it tackles complex industrial problems such as routing, scheduling and real-time computer vision. For the Aisin Group, it cut the number of trucks needed between warehouses by 30% and achieved a 53% cost reduction.
Live Models
Tiny AI models built on the fly that learn and act at the same time.
Conventional AI trains one large model in advance, then deploys it to make predictions. Funartech’s Live Models work differently. We create very small, purpose-built models on the fly, while the problem is being solved, and they keep training continuously as they are used, so training and inference happen at the same time. This removes the classic split between learning and using.
Live Models adapt instantly to the situation in front of them and need little data and little computing power. They can run where large models can’t, such as on edge devices with no internet connection. When a human expert does step in and a full retrain is needed, it costs very little because the models are so small. The result is fast, frugal and adaptive AI that stays relevant as conditions change.
Know-How-Augmented AI
AI built on data, scientific knowledge and human expertise, not on data alone.
Data alone is often incomplete, biased or of poor quality, and AI that relies only on data inherits those weaknesses. Funartech’s solutions combine three sources. The first is data. The second is formal knowledge: the physics, rules and constraints of the problem. The third is the practical know-how of our clients’ experts, captured and built into the models.
Know-How-Augmented AI delivers solutions that respect real-world constraints and remain reliable when data is scarce or messy. They are explainable to the people who use them, so they are trusted and adopted on the ground, not just accurate in the lab. It builds on our hybridization of ML and OR.